Stanley Chow
Mathematics @ HKU · Second Major in Computer Science · Minor in Finance
Recommendation Systems · Search & Ranking · Machine Learning · Applied Mathematics
I build retrieval and ranking pipelines, reproducible machine-learning experiments, and numerical methods for inverse problems. My current interests center on recommendation, search, and advertising algorithms, with mathematical modeling and scientific computing as a complementary research foundation.
Recent work includes a nine-channel H&M recommendation pipeline with learning-to-rank and temporally frozen evaluation, a deterministic C++ retrieval layer for document question answering, and HKU research on visible-geometry recovery and coverage-aware reduced-order models for parabolic inverse problems.
View projects Download CV GitHub
Selected Work
H&M Two-Stage Personalized Recommendation System
An offline retrieval-and-ranking system built on 31.8 million H&M transaction events.
- Combined nine heuristic, collaborative, learned, text, and image retrieval channels in a quota-aware candidate interface.
- Trained a LambdaRank model on 100,000 customer-week queries under four rolling temporal cutoffs; an untouched future week reached MAP@12 of 0.03448.
- Ran bounded-memory inference for 1,371,980 customers across 138 shards, with Kaggle scores of 0.03117 public and 0.03116 private MAP@12.
Parabolic Inverse Problems & Coverage-Aware POD
Ongoing HKU Summer Research Fellowship work on recovering source geometry from noisy diffusion observations and building reduced models that remain reliable beyond their snapshot law.
- Separated raw source error from heat-visible and geometric recovery.
- Developed covariance-designed POD with held-out, gap-free risk diagnostics.
- Implemented finite-difference, Tikhonov, Monte Carlo, and reduced-order experiments.
DocuQuest Agent
A retrieval-augmented document question-answering system with a deterministic C++ search layer.
- Implemented tokenization, a custom ownership-aware BST multimap, and an inverted index.
- Applied AND constraints within expanded term groups and unions across groups.
- Kept retrieval tests deterministic and credential-free through an injected model client.
Behavioral Personality Analytics
A leakage-resistant study of personality and behavioral-outcome prediction.
- Evaluated 381 model-and-feature-subset configurations using development data only.
- Selected a compact three-feature Gradient Boosting model with 0.9651 holdout ROC-AUC.
- Reported limitations for generated data, external validity, and causal interpretation.
Technical Toolkit
Languages: Python · C++ · SQL · MATLAB
Machine Learning & Data: PyTorch · LightGBM · XGBoost · scikit-learn · pandas · NumPy · SciPy · DuckDB
Recommendation & Search: Candidate Retrieval · Collaborative Filtering · Two-Tower Models · LightGCN · LambdaRank · Negative Sampling · Temporal Validation
Research & Engineering: Git · CMake · Jupyter · LaTeX · Numerical Optimization · PDEs · Experiment Design
I am open to internship and research opportunities in recommendation, search, ranking, advertising algorithms, machine learning engineering, and applied data science.